Research On The Fault Diagnosis Method Of Rolling Bearing Of AC Motor Based On Data Fusion

With the rapid development of power electronics technology,AC motors in machine tools,compressors,new energy vehicles and other fields of application gradually expanded,motor health or not directly affect the stability and safety of the entire transmission system.Rolling bearing as a key component of AC motor operation,its failure will directly affect the normal operation of the AC motor,and according to relevant statistics,about 44% of the motor failure is caused by bearing failure.Therefore,it is important to detect and troubleshoot the operating condition of motor bearings.In daily operation of motors,the bearing fault signature component of the motor current signal shows time-varying non-mooth characteristics due to load variations,which makes it difficult for Motor Current Signature Analysis(MCSA)to extract the bearing fault signature component and make fault diagnosis of motor rolling bearings smoothly;The infrared image,however,can have overlapping background heat such as bearings and rotors,which can cause confusion in identifying healthy bearings with faulty bearings.In view of the above problems,in order to improve the accuracy of rolling bearing fault diagnosis when the motor is operated under non-smooth working conditions.In this paper,the current signal and infrared image signal are used as the parameters for bearing fault diagnosis,and an AC motor rolling bearing fault diagnosis method based on the fusion of current and infrared image heterogeneous data is proposed.The main research of this paper is as follows.To solve the problem that the current signal is too complicated and the model running time is too long due to the large amount of data,VMD is used to decompose the motor current signal,and the most suitable K value is selected to decompose the signal according to the principle of maximum correlation coefficient to extract the low frequency band component where the bearing fault signal is located;After that,the current signal of low frequency band is transformed into a two-dimensional map suitable for convolutional neural network training by using the two-dimensional formula of one-dimensional data,and the convolutional neural network is trained,and finally the trained convolutional neural network and softmax classifier are used for fault classification of the current data set.Pre-processing of infrared images is required because they are subject to various complex and useless information when using infrared images for fault diagnosis of motor bearings.Firstly,the grayscale image is grayscale transformed to obtain the grayscale map,then the grayscale map is image segmented and the grayscale feature matrix is extracted,so that the similarity is calculated with the grayscale feature matrix of the infrared image of the faulty bearing in the fault library,and finally the sigmod classifier is used to classify the similarity data to make a judgment on the infrared image diagnosis result.A decision-level fusion diagnostic model is introduced to fuse the outputs of the current diagnostic model and the infrared image diagnostic model by weight,and the fusion model is trained to perform fault diagnosis of motor bearings by the completed fusion model.In order to verify the effectiveness of the fault diagnosis method based on heterogeneous data fusion in this paper,an experimental platform for variable load motor bearing fault diagnosis is built,and the motor current signals and corresponding infrared image signals are collected and analyzed and processed to experimentally verify the method in this paper,followed by analysis and comparison with different diagnosis methods using the same data set.The experimental results and comparison results show that the fault diagnosis method of heterogeneous data fusion in this paper can diagnose the fault of motor bearing outer ring under the load variation,and the diagnosis effect is the best,and the fault diagnosis accuracy can reach 98.85%.